Re: [PATCH] tst-realpath-toolong: Fix hurd build

[Samuel Thibault <samuel.thibault@aquilenet.fr>]

Siddhesh Poyarekar, le dim. 23 janv. 2022 20:49:48 +0530, a ecrit:
> On 23/01/2022 06:06, Samuel Thibault wrote:
> > Siddhesh Poyarekar via Libc-alpha, le sam. 22 janv. 2022 20:15:23 +0530, a ecrit:
> > > We don't really need a bigger buffer for realpath since it should fail
> > > and return NULL.  In the bug too, the buffer itself is not accessed; it
> > > is in fact left untouched.  Drop the PATH_MAX use and pass a single char
> > > address.
> > 
> > ? realpath assumes that the passed buffer is PATH_MAX-long. When
> > PATH_MAX is not defined, calling it with a buffer is essentially
> > undefined. Better just pass NULL.
> 
> Passing NULL doesn't reproduce the problem because realpath just allocates
> enough to accommodate the return, even when it exceeds PATH_MAX.

Ah, ok, sorry I didn't check the whole story.

> Would you prefer it if I defined PATH_MAX on hurd then, something like:
> 
> #ifndef PATH_MAX
> # define PATH_MAX 1024
> #endif
> 
> or do you prefer a more accurate path_max value using pathconf()?

Better use the accurate from pathconf() ; that being said it will return
-1 on the ext2fs filesystem, so we'll have to resort to a hardcoded
limit in that case anyway. I'm fine with just setting PATH_MAX by hand
here.

Samuel